Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75052.87020858056,
-      "required_pages_per_second_upper_estimate": 0.006505392569919293
+      "required_token_positions_per_second": 75092.67453620704,
+      "required_pages_per_second_upper_estimate": 0.00650884270815481
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9381.60877607257,
-      "required_pages_per_second_upper_estimate": 0.00374246400832638
+      "required_token_positions_per_second": 9386.58431702588,
+      "required_pages_per_second_upper_estimate": 0.003744448826003622
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18763.21755214514,
-      "required_pages_per_second_upper_estimate": 0.00434054015933832
+      "required_token_positions_per_second": 18773.16863405176,
+      "required_pages_per_second_upper_estimate": 0.004342842167004356
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28144.82632821771,
-      "required_pages_per_second_upper_estimate": 0.0017138722588968276
+      "required_token_positions_per_second": 28159.752951077637,
+      "required_pages_per_second_upper_estimate": 0.0017147812119150594
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28144.82632821771,
-      "required_pages_per_second_upper_estimate": 0.002166666640611929
+      "required_token_positions_per_second": 28159.752951077637,
+      "required_pages_per_second_upper_estimate": 0.0021678157333592228
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28144.82632821771,
-      "required_pages_per_second_upper_estimate": 0.3494645483221092
+      "required_token_positions_per_second": 28159.752951077637,
+      "required_pages_per_second_upper_estimate": 0.34964988702183636
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.42963884407802766,
-    "producer_p95_seconds": 0.21847197800161666,
-    "gpu_update_seconds": 0.9823475077648783,
-    "producer_batches_per_second": 2.327536287241262,
-    "gpu_batches_per_second": 1.0179697022648189,
-    "producer_headroom": 2.2864494710037717,
+    "producer_batch_seconds": 0.08520502524220319,
+    "producer_p95_seconds": 0.22251638599846046,
+    "gpu_update_seconds": 0.9818267954279744,
+    "producer_batches_per_second": 11.736396969045044,
+    "gpu_batches_per_second": 1.0185095830106206,
+    "producer_headroom": 11.523109025988086,
     "sustainable": true,
-    "buffer_seconds": 60.90554548142246,
+    "buffer_seconds": 60.87326131653441,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 254674.87845545448,
-    "upload_bytes_per_second": 5165927.2819828605,
+    "generation_bytes_per_second": 252398.75749624867,
+    "upload_bytes_per_second": 5357611.107334302,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.64730038879133,
+    "estimated_full_shards_per_hour": 54.15889781632991,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.532518773999982,
-      "bytes": 2518854,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2675222210000356,
       "bytes": 11869039,
@@ -562,6 +556,12 @@
       "bytes": 2109863,
       "pages": 20,
       "shards": 3
+    },
+    {
+      "seconds": 1.3331431920014438,
+      "bytes": 16665888,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -189,3 +189,4 @@
 {"step": 213900, "loss": 0.5893921442329884, "lr": 0.0003, "grad_norm": 0.14274787417811938, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31307863405, "tokens_per_second": 149522.80302397424, "data_wait_s": 1.0782998288050294e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214000, "loss": 0.6256532184779644, "lr": 0.0003, "grad_norm": 0.16470994345720258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31322486539, "tokens_per_second": 124167.84889750848, "data_wait_s": 1.074199826689437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214100, "loss": 0.6236763224005699, "lr": 0.0003, "grad_norm": 0.15930699896836142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31337109524, "tokens_per_second": 149078.56328341994, "data_wait_s": 1.6582001990173012e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 214200, "loss": 0.5742165800184011, "lr": 0.0003, "grad_norm": 0.13650238462357717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31351757875, "tokens_per_second": 149042.78802969292, "data_wait_s": 1.4353998267324641e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:509f1b1211ab97dfbd10bad66ee979a6c94bb92cb779ea6fc2b1554931102adf
-size 407052
+oid sha256:460fe240ccab3c4e6f5063d18235c2eb8117067fefc3e67f182239fc0618d8ab
+size 409170